SGI: Discussion

Did anyone else see this already? Read the paragraph at the bottom.

http://www.theregister.co.uk/2011/02/25 ... page2.html


Simon

It uses those blasted DDR2 mini-DIMMs. So that's (possibly) why my shop shipped almost 1000 sticks to china recently.

Anyways, that was suprisingly unexpected and refreshing.

I wonder how much it would take to port IRIX to that architecture? From what I can see, it's vanilla enough as a CPU - effectively it's an R16k@1GHz but with 8 cores. I should imagine that the fact IRIX can natively support NUMA and single-image computing over multiple nodes, I'd think that this could be patched in relatively easily.

I don't know what the pin array is, but it could conceivably be mounted on a daughterboard for inclusion in machines like the Onyx2, Tezro and Octane, as they would probably have enough clearance - much like the old 68040/68060 accelerator boards for Amigas and 68k Macs. Now that would be interesting - especially as the state-of-the-art PCs still struggle to maintain the system-wide bandwidth of something like an Onyx2 or even an O2 - they rely heavily on multiple cores and high clock speeds to overcome the bottlenecks through brute force, but a quick (1-2ghz) multicore MIPS CPU fitted into a proper architecture would take less power, do as much actual work on a system-to-system comparison and would work with things like Flint, Flame and Smoke as well as all the other old heavyweight apps that are being sold off at fire-sale prices...

I can dream...!

chancehooper wrote: I wonder how much it would take to port IRIX to that architecture?

Nobody left at SGI that knows anything about IRIX anymore

Nothing is impossible to reverse engineer, but also if this became a viable platform to license to, I'm sure they could find people to de-compile the code they have, not to mention re-hiring staff.

Having said that, even the hobbyist developer communities can pretty much crack into any kernel or sealed code-base - as long as they got agreement from SGI/rackable. Which is feasible as they have no commercial interest in it, even to the point of ending MIPS/Irix support. Who knows? As I stated before, this is more a wish-list than a practical wel-researched proposal!

pentium wrote: It uses those blasted DDR2 mini-DIMMs. So that's (possibly) why my shop shipped almost 1000 sticks to china recently.

the article says DDR3.

kshuff wrote:

chancehooper wrote: I wonder how much it would take to port IRIX to that architecture?

Nobody left at SGI that knows anything about IRIX anymore

agreed. that horse has been beaten to death a hundred times over on the board. please no more, nobody has anything new to say.

chancehooper wrote: ...as long as they got agreement from SGI/rackable. Which is feasible as they have no commercial interest in it, even to the point of ending MIPS/Irix support.

${same_old_boilerplate_about_sanitizing_the_code_and_finding_out_SGIs_licensors_costing_money_that_Ive_said_before}

fieldframe wrote: Did anyone else see this already? Read the paragraph at the bottom.

http://www.theregister.co.uk/2011/02/25 ... page2.html


Simon

Too bad Hamei's gone - it would be nice to see the inside scoop. I don't think he was ever able to track down one of the Godson things, was he?

Pretty comical that they mentioned Sgi ^^

' Wouldn't it be funny if Silicon Graphics started building systems with these Godson-3 chips?
They could dust off Irix and take it out for a spin on some new iron and allow it to run x64-based
Linux applications in emulation mode. ® '

SGI was the major former MIPS maker. The association would be natural. All the same SGI is highly unlikely to consider it. It would require retooling of their ASICS, dusting off MIPSpro and bringing it up to speed with the new chips, and you still have the uncertainty of any Chinese logic in there - the Chinese don't really have the best reputation for overdesigning and extensive testing, and little bugs get expensive very fast in SGI's market. Their buyers don't buy the old "it will be fixed in a later firmware update which may or may not actually happen" or the intimation that it's the buyers fault and they need to buy a new machine. How many millions did the original R10k fiasco cost SGI again?

It'll be interesting to see if anybody besides the Chinese researchers can get their hands on the Godson chips in quantity - or did I miss an announcement somewhere? Last I knew was Hamei's lament that he couldn't find any in-country.

When somebody has shipped viable machines running Red Flag Linux (or whatever) outside of China, maybe people will start paying attention.

smj wrote: When somebody has shipped viable machines running Red Flag Linux (or whatever) outside of China, maybe people will start paying attention.

I've seen brand new laptops with RedFlag Linux (x86) pre-installed in Malaysia. The sales clerk then said "of course you'll want Vista.." before I walked away, it probably would not have fit where I was about to stick that laptop and Vista and he sure as hell wouldn't have appreciated it.

kshuff wrote:

chancehooper wrote: I wonder how much it would take to port IRIX to that architecture?

Nobody left at SGI that knows anything about IRIX anymore

Not only that, but they act like they hate MIPS and IRIX.... Or at least the local (Australian) office seems find it an offensive topic to bring up.

SAQ wrote: To bad Hamei's gone - it would be nice to see the inside scoop. I don't think he was ever able to track down one of the Godson things, was he?

Hamei did come back briefly under a different name, made a few posts, and was run off again..

SAQ wrote: SGI was the major former MIPS maker. The association would be natural.

Not only that but they owned MIPS for a time. The modern era SGI is a completely different animal, with as much in common with the old SGI as say Disney, Mattel or Dairy Queen.

R.

PymbleSoftware wrote: Not only that, but they act like they hate MIPS and IRIX.... Or at least the local (Australian) office seems find it an offensive topic to bring up.

R.

I imagine that it is probably something like a singer or sports player when someone keeps dwelling on the past "glory days" and they're wishing someone would notice what they were doing now. Especially because the "now" stuff is what's paying their salaries.

SAQ wrote: To bad Hamei's gone - it would be nice to see the inside scoop.

I'm sorry, gone? Did I miss something?

chancehooper wrote: ... especially as the state-of-the-art PCs still struggle to maintain the system-wide bandwidth of something like an Onyx2 or even an O2 - they rely heavily on multiple cores and high clock speeds to overcome the bottlenecks through brute force, but a quick (1-2ghz) multicore MIPS CPU fitted into a proper architecture would take less power, do as much actual work on a system-to-system comparison and would work with things like Flint, Flame and Smoke ...

It would be nice to believe this is true and be correct, but alas it's not, by a long way. O2's STREAM results are horrible, while the limited
mem/gfx bw of IR and VPro mean that neither can do more than just over 8-bit 2K. Not sure if IR4 improved the bw rates. Discreet never
supported multiple pipes, and also did not support VPro in high-end systems. In other words, CPU stuff aside, it needs modern gfx to do
this properly. By contrast, modern x86 can handle 4K (6 bytes per pixel) no problem. My "state of the art" PC easily matches/beats the I/O
of a Tezro (in quotes because my PC isn't state of the art, it's just a single quad-core i7 870, tested with just one SAS card and 4 disks,
ie. nowhere near the potential of a dual-socket X58 system with 2 x 6-core Gulftown, etc.)

Of course one can get round the limited gfx bw of IRx for IFFFS by using proxies, but that's a whole other ball game.

As for IRIX ever coming back, other posters are right, nice to dream but it ain't gonna happen. Collective effort would be better spent on
updating freeware apps, some of which are woefully out of date now, eg. mplayer, and of course sorting out browser ports.

Ian.

Technically PCs have come a long way. Hypertransport, PCIe, and well-done SATA have made life much easier - though it's true that these are rarely done properly on the el-cheapo models.

I knew things were getting better when I could finally do something disk-intensive and have the computer still be responsive at the same time. Note that this doesn't mean that things can't be done better, just that the state-of-the-art in "cheap but good enough" has risen to where it is not bad at all.

Hamei did come back briefly under a different name, made a few posts, and was run off again..

So you're saying bluebird on a branch was Hamei? I had no idea...

vishnu wrote:

Hamei did come back briefly under a different name, made a few posts, and was run off again..

So you're saying bluebird on a branch was Hamei? I had no idea...

Walks like Hamei, talks like Hamei, does the west-coast-hippie burning-man chants in a repressed regime like Hamei... No, I really don't have any evidence other than the lyrics to that song in b-o-b's posts... the kinda thing only he would wax lyrical.... If you really knew Hamei, you'd recognize his style.

R.

Godson/Loongson 2F was available in a variety of commercial netbooks, most notably the French Gdium and Chinese Lemote Yeeloong (along with some Leemote net-tops). It has strange hardware bugs, notably surrounding branch prediction (pretty important, I'd think), and the errata wasn't translated into the English manual (!!!).

Stallman carried a Lemote netbook for a while as the entire hardware/software stack from the ground-up was open-spec and open-source. Sadly they were also garbage.

bri3d wrote: Stallman carried a Lemote netbook for a while as the entire hardware/software stack from the ground-up was open-spec and open-source.

that's good enough for me to pass on it.

bri3d wrote: Sadly they were also garbage.

why am i surprised?
i'm not really.